RESCAST-100K: A Comprehensive Dataset for Cross-Domain Residential Load and Indoor Temperature Forecasting

In three linesRESCAST-100K is a benchmark of 100,000 U.S. homes simulated via EnergyPlus/ResStock for evaluating cross-domain generalization in residential energy load and indoor temperature forecasting. 15-minute time series dataset with 40+ static building covariates, integrating 5 real-world datasets. Cross-attention and MLP-mixer models outperform classical transformers under domain shift.

Summary generated by Claude — human-verified